Has anyone evaluated how much stack-v3-train improves downstream coding models compared to stack-v2? I'm especially interested in HumanEval, LiveCodeBench, or SWE-bench results.
· Sign up or log in to comment
